ResearchBenchmarksAutoregressive Generation on Long-context evaluation document 61,651 tokensFollow26.4Time to First Token (ms)Nectar11.936109.568207.2304.832May 10, 2026Evaluation ResultsMethodMethodLinksTime to First Token (ms)Tokens per SecondMemory Usage (GB)NectarModel=Qwen3-1.7B, Infe...Model=Qwen3-1.7B, Inference Protocol=Nectar2026.0526.442.97.9NectarModel=Qwen3-4B, Infere...Model=Qwen3-4B, Inference Protocol=Nectar2026.054717.516.8BaseModel=Qwen3-1.7B, Infe...Model=Qwen3-1.7B, Inference Protocol=Base2026.052147.334.2BaseModel=Qwen3-4B, Infere...Model=Qwen3-4B, Inference Protocol=Base2026.05388450.6